OpenAI has enhanced security measures to protect its AI developments from industrial espionage and copying by competitors. According to Financial Times, the company accelerated the implementation of new policies after the Chinese startup DeepSeek released its own model in January, and OpenAI accused DeepSeek of using “distillation” methods for unauthorized copying of its developments.
The company restricted employee access to internal algorithms and new products by implementing an “information tenting” policy. During the development of the “o1” model, only vetted team members with the appropriate clearance were allowed to discuss the project in office premises.
OpenAI moved some technologies to offline computers, implemented biometric access control to office areas using fingerprint scanning, and established a “deny-by-default” policy for internet connections, requiring separate permission for external connections.
Simultaneously, the company increased the physical security of data centers and expanded its cybersecurity staff, including hiring Dane Stucky and retired U.S. General Paul Nakasone.
